A short video clip from a 2023 Senate hearing has gone viral again thanks to a post by CoinBureau. In it, Anthropic CEO Dario Amodei warns lawmakers that the scaling of open source AI models is heading down a “very dangerous path.” The post has already pulled in hundreds of thousands of views and thousands of replies, with people arguing over safety, profits, innovation, and who should control powerful technology.
Amodei made the comments during a July 2023 hearing before the Senate Judiciary Subcommittee on Privacy, Technology, and the Law. He testified alongside AI researchers Yoshua Bengio and Stuart Russell. The focus was on principles for AI regulation and the risks that come with rapid progress.
He started by saying open source has value in most scientific fields and even in AI for smaller or medium-sized models. Those carry limited risks, he noted, and the benefits often outweigh them. But he drew a sharp line when it came to the most powerful frontier models.
“If we talk about two to three years for the frontier models… I think the path that things are going in terms of the scaling of open source models, I think it’s going down a very dangerous path,” Amodei said. He explained why. Once a company releases the full model weights openly, it loses the ability to monitor how people use it, revoke access if someone misuses it, or push safety updates to fix problems that emerge later.
The model is simply out there. Anyone can download it, fine-tune it, or run it without any oversight from the original creators. That loss of control, he argued, creates real problems for preventing misuse in areas like biology, cybersecurity, or misinformation.
Why the clip is back in the spotlight now
The CoinBureau post shared the clip on June 28, 2026, and it quickly gained traction. Many viewers saw it as a timely reminder of long-standing tensions in the AI world. Others treated it as proof that big closed-model companies want to slow down competition.
Anthropic built its reputation around safety. The company describes itself as a public benefit corporation focused on making AI systems more controllable and less likely to cause harm. Its Claude models come with guardrails designed to reduce risks around misuse. Critics on social media quickly pointed out that Anthropic also makes money by charging for access to those same models through APIs.
If truly powerful open source alternatives spread widely and for free, that paid model could face pressure. Some replies called the warning self-serving. Others compared it to older tech battles, like concerns raised years ago about open source operating systems challenging established players.
The state of open source AI in 2026
The debate lands differently today than it did in 2023. Open source models have improved dramatically. Projects from Meta with its Llama series, along with models out of Chinese labs like DeepSeek and Qwen, now deliver performance that sits close to or matches many closed frontier systems on standard benchmarks.
Users can often run these models locally or through cheaper hosting options. Costs drop significantly compared to paying per token through commercial APIs. Companies and developers gain more flexibility to customize the models, audit the code themselves, and avoid depending on a single provider that might change terms or pricing later.
Closed models from Anthropic, OpenAI, and Google still lead in some complex reasoning tasks and come with polished safety features out of the box. But the gap has narrowed enough that many teams now mix and match. They use open models for cost-sensitive workloads and closed ones only when they need the absolute top performance or built-in compliance tools.
China has pushed hard on open releases in some cases, which adds a national security angle to the discussion. Amodei and other Western leaders have raised concerns in recent months about how open models could accelerate capabilities for adversaries without the same safety checks.
Arguments on both sides
Supporters of stricter controls on powerful open releases say the risks are real and growing. A model capable of helping with advanced biology or sophisticated cyber attacks becomes far harder to contain once the weights are public. Bad actors could fine-tune it for harmful purposes without the original developers ever knowing. Updating safety features after release becomes nearly impossible.
They point to the pace of progress. What seems manageable today could look different in another year or two as models get stronger.
More Read
On the other side, advocates for open source argue that transparency and broad access drive faster innovation and prevent any single company from holding too much power. Community scrutiny can catch problems that internal teams miss. Local or self-hosted models give users more privacy and control over their data. Many see calls for heavy restrictions as a way to protect existing business models rather than pure safety concerns.
History offers examples on both sides. Open source software transformed computing and created enormous value through collaboration. At the same time, certain technologies carry inherent dangers that societies have chosen to regulate tightly.
What happens next
The resurfaced clip highlights a core tension that regulators, companies, and developers will keep wrestling with. How do you encourage rapid progress and wide access while reducing the chance that powerful tools fall into the wrong hands or get used in ways that cause serious harm?
Some push for targeted rules around the most capable models, such as mandatory testing before wide release or licensing requirements for frontier systems. Others want lighter touch approaches that focus on misuse after the fact rather than limiting development upfront.
Users and builders continue voting with their choices. Many experiment with open models for everyday work while keeping an eye on safety tools and best practices. The conversation shows no sign of slowing down.
Amodei’s three-year-old warning still resonates because the underlying questions remain unsettled. Powerful AI is advancing quickly. Whether open or closed approaches ultimately prove safer and more beneficial depends on how the industry, governments, and communities handle the trade-offs ahead. The viral clip simply brought those questions back into sharp focus for a new audience.
(Word count: 1,012)
